File Exchange

image thumbnail

Fast/Robust Template Matching

version 1.2 (13.4 KB) by

Template Matching with SSD Block Matching and Normalized CC, (color) 2D and 3D



View License

TEMPLATE_MATCHING is a CPU efficient function which calculates matching score images between template and (color) 2D image or 3D image volume.

It calculates:
- The sum of squared difference (SSD Block Matching), robust template matching.
- The normalized cross correlation (NCC), independent of illumination, only dependent on texture

The user can combine the two images, to get template matching which works robust with his application.

Both matching methods are implemented using FFT based correlation.

Try the examples.

Please report: bugs, successes and other comment.

Comments and Ratings (30)


Zarrin (view profile)

Thanks for sharing! I want to use this code to develop 3D correlation. However, when I check your code with 2D image and compare the results with 'normxcorr2' code, they are different. And I have checked 'normxcorr2' gives the right result.


shaojiang wu

Does this code detect the template which is slightly deformed from the original image?

huang samuel

very fast , i liked it


How can I allow in the 3D case the matching only in the z-direction (and so prohibit the matching the x- and y-direction)?


jiajia (view profile)


Excellent Code. Thank you.
I use it to find ellipses in a 3d image. The shape and the orientation of the ellipses are not exact equal. Is it possible to find ellipses which are similar (shape and orientation) to the template?


Jem (view profile)

help me :(


2one (view profile)

Excellent code but slight issue if you have a M x N x 3 3D grayscale image volume then the line:


will cause a problem. Suggested edit:

if(size(T,3)==3) & (size(I,3)==3)

Excellent work -- used this to track a blurry sphere in a video sequence.

Ilya Belevich


Mai (view profile)

Is the 2D version of the normalized cross-correlation in this function the size as normxcorr2? If you have a paper describing this method, I would like to have it! Thank you!



Thanks so much. I was wondering how to change the code so it would detect more than one match Using Idata which I seem clueless on how to set it.
Thanks in advance

pham bao

thanks for share. But I have a problem: "??? Maximum recursion limit of 500 reached. Use
to change the limit. Be aware that exceeding your available stack
space can
crash MATLAB and/or your computer.

Error in ==> imbmpinfo>decodeCompression"
I'm using Matlab 7.8.0(R2009a)
Can you help me !

Does anyone know which paper or algorithm this code use?


Shashi (view profile)

Please indicate the paper you used to implement this. And added to it, if the template size is smaller than the image size, what changes are to be made in the code.


Please indicate that it requires the image processing toolbox or avoid using functions from the toolbox. padarray is used here and it's from the im.proc. toolbox.


Carlos (view profile)

Hi, You used any paper to develop this code? Can you give me the name of this paper?

Thanks a lot

The problem is that z is logical and I want it to be the number of the slice.

Hi, I have a problem with the NCCC three-dimensional, with the code I have written above, it plots a * in all the slices not only in the slices that match the template.
What should I do? Thanks in advance.

Hi, thanks for sharing your code, it's great. I have a 3D image with a sphere, and another sphere as a template. To locate the maximum cross correlation I write:
daspect([1 1 1]);
figure,isosurface(I,0); hold on; plot3(y,x,z,'r*'); title('MAX NCCC3D')

Is it correct?
I don't know if the real coordinates are displayed well because I don't understant very well the normalized cross-correlation in 3D.

Thanks very much.


Thanx for sharing your code. How can I detect same template apping more than once in image.


I tried only the 3D normalized cross-correlation. It works very well. Thanks for sharing.

Dirk-Jan Kroon

Dirk-Jan Kroon (view profile)

The submitted function supports 3D, SSD and color which is not in the function 'normxcorr2'. Also normxcorr2 is only a little bit faster when the template is only a few pixels in size...


Tony (view profile)

The function 'normxcorr2' available in matlab is faster than the code you submitted. Please profile it and see for yourself. But thanks for sharing the code.


Masoud (view profile)

Perfect. It is fast and robust.
Thank you



Faster Multiple Template search


Fixed Range to 0..1

MATLAB Release
MATLAB 7.8 (R2009a)

Download apps, toolboxes, and other File Exchange content using Add-On Explorer in MATLAB.

» Watch video